Seeing through Unclear Glass: Occlusion Removal with One Shot

Qiang Li; Yuanming Cao

REVIEW 4 major objections 5 minor 50 references

Seeing through Unclear Glass: Occlusion Removal with One Shot

T0 review · 4 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper claims that one all-in-one network can remove dirt, raindrops, muddy water, and particles from a single photo taken through contaminated glass, adapting itself to the unseen contaminant type at test time.

desk verdict A genuinely useful dataset and a plausible test-time-adaptation baseline, but the headline gains are undercut by test-set tuning of the update count and an internal inconsistency about that number. read the letter →

arxiv 2509.01033 v1 pith:3224XVKD submitted 2025-09-01 cs.CV cs.AI

Qiang Li , Yuanming Cao This is my paper

classification cs.CVcs.AI

keywords occlusionremovaltest-timeadaptationself-supervisedauxiliarylearningunclearglassimagerestorationrealpaireddatasetdefocusdeblurringinpainting

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper's goal is to show that a photograph taken through glass covered with dirt, raindrops, muddy water, or small particles can be restored from that single image, without hardware changes and without a separate model per contaminant. The route is a real paired dataset of degraded and clean shots, called OROS, plus a two-branch neural network that learns two things at once: removing the occlusion and reconstructing it. The second, self-supervised task lets the model spend a few gradient steps on each new test image so it adapts to that image's particular occlusion—this is the one-shot mechanism. If the claim holds, the practical consequence is that surveillance, dashcam, and through-window photography could be cleaned on the fly with one model and no per-type retraining.

What carries the argument

The load-bearing identity is the unified occlusion expression I = IN + (α − 1)·IP + β·IC, which maps the two physical regimes—partial occlusion (defocus-blurred margins) and complete occlusion (underexposed cores)—onto the familiar restoration tasks of deblurring and inpainting. The other carriers are the OROS real-pair dataset, collected with two identical glasses on a tripod and remote control to keep the scene static, and the two-branch autoencoder whose auxiliary self-reconstruction branch produces an occlusion attention mask and enables per-image gradient updates during testing.

What would settle it

Measure the residual registration error between the contaminated and clean images in the OROS pairs (e.g., dense optical flow on occlusion-free regions). If the residual misalignment exceeds about one pixel, or if the PSNR margin over DeblurGAN-v2 evaporates after alignment-robust evaluation, the supervised-training claim is weakened. A second check: run the test-time adaptation with 0 vs 20 updates on a fully held-out contaminant class and compare human-rated restoration; the one-shot claim requires the adapted output to be clearly better than no update.

Watch

Extended reading notes

Core claim

The central claim is that occlusion removal through a marginally distant unclear glass can be treated as one unified task: the degraded image decomposes as I = IN + (α−1)·IP + β·IC, where partial occlusion behaves like defocus deblurring and complete occlusion like inpainting. Built on this, the paper proposes an all-in-one network with a primary restoration branch and a self-supervised auxiliary branch that reconstructs the contaminated input; at test time, a few gradient steps on the auxiliary loss adapt the shared weights to the unique occlusion of that single image. On the 333-image OROS test set it reports PSNR 29.91 dB and SSIM 0.879, ahead of re-trained baselines including DeblurGAN-v

Load-bearing premise

The load-bearing premise is that every degraded/clean pair is aligned well enough that the supervising loss points at the same scene; the paper does not quantify the residual misalignment left after using two identical glasses and a tripod.

Editorial extensions

If this is right

A single trained model, updated briefly on the test image itself, can handle contaminant types it never saw during training; the paper's held-out Dirt experiment shows adaptation raising PSNR from 29.39 to 29.67 dB with 3 updates.
Occlusion removal reduces to the combined tasks of defocus deblurring and inpainting, so any architecture that solves those two well can be assembled along the same unified degradation expression.
Real image pairs, not synthetic ones, are a viable and sufficient training signal for this task; the paper reports that all baselines retrained on OROS still trail the proposed model.
The one-shot update can be folded into deployment at modest cost: 1.30 s per 256×256 image with 6 gradient updates on a 3090 GPU, versus 1.68 s for DeblurGAN-v2.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Editorial: the same test-time adaptation scheme should transfer to other degradations—haze, lens smudges, curved-windshield raindrops—because the auxiliary task only needs the degraded image to define its loss.
Editorial: the unified expression predicts that any contaminant following attenuation-plus-intensification will yield to the same per-image update; collecting pairs for a new, unlisted occluder is a direct way to test that prediction.
Editorial: since the dataset pairs were captured with two separate glasses, the reported margin could be partly influenced by residual misalignment on thin, textured dirt; an experiment measuring registration error against PSNR gains would clarify how much of the margin is true restoration.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Here's the take: the OROS dataset is the real contribution. Real paired clean/contaminated photos through glass for four occlusion types, collected with a tripod and two identical glasses. That's work, and the field can use it. The method is a reasonable combination of existing pieces: multi-scale U-net, auxiliary self-reconstruction, occlusion attention mask, and test-time adaptation. Nothing here is a new principle, but doing it for mixed contaminants in one model is a legitimate extension.

The empirical claim is shakier than the abstract suggests. The stress-test note lands: the paper tunes the number of gradient updates on the test set and reports it inconsistently. Section 4.1 says 6 updates; Section 4.3 says best n=4 and Fig. 9 shows n=4; Table 3 says they implemented 20 updates and got best at 3. That is not a minor typo. The 'unseen ones' table, which is central to the 'especially unseen ones' claim, is explicitly reporting the test-set-optimal number of updates. Without a held-out validation split for choosing n, the reported margin over DeblurGAN-v2 is not a faithful estimate of generalization.

Other soft spots: Eq. 3 is more re-labeling than derivation; there are no error bars, so we don't know if the PSNR gaps matter; and manual selection of 'defective' pairs could bias the dataset, though that's common practice and not disqualifying. Also, the text promises to open-source code and dataset, but no artifacts are linked. Given the claims, that should be a condition of acceptance.

Net: this is a legitimate empirical paper with a useful dataset and a sound general approach, but the headline numbers need to be redone with a disciplined protocol: fix n by validation, report variance, release the data. I'd send it to review rather than desk-reject, but only with the expectation that the authors address the update-count issue. The paper is worth a serious referee's time because the dataset alone is valuable, and test-time adaptation for this problem is a plausible direction that deserves scrutiny.

Referee Report

4 major / 5 minor

Summary. The paper introduces OROS, a new real paired dataset of images captured through glass contaminated by dirt, raindrops, muddy water, and particles, and proposes an all-in-one occlusion removal network. The method uses a two-branch architecture: a primary branch for clean image prediction and a self-supervised auxiliary branch that reconstructs the input degraded image. At test time, the auxiliary reconstruction loss is used to adapt the shared parameters to each test image. The authors claim the method outperforms state-of-the-art methods on OROS, especially on unseen occlusion types.

Significance. The OROS dataset is a potentially valuable resource, as real paired contaminated/clean images are rare. The test-time adaptation idea, using a self-supervised reconstruction task, is well motivated and could improve generalization to unseen occlusions. However, the theoretical degradation model is presented as a contribution but is not rigorously derived, and the experimental evaluation suffers from test-set hyperparameter selection, missing error bars, and unclear dataset alignment. If these issues are addressed, the work would be a useful empirical contribution.

major comments (4)

[§4.1, §4.3, Table 3, Fig. 9] The number of test-time gradient updates is selected on the test set and is reported inconsistently. §4.1 says 'We perform 6 gradient updates for each test image,' while §4.3 and Fig. 9 report the best performance at n=4, and Table 4 (which reaches the headline PSNR 29.91 dB, same as Table 1) uses n=4. In Table 3, the authors state 'We implement 20 updates and achieve the best quantitative performance in 3 updates.' This constitutes test-set tuning. Since the claimed advantage over SOTA, especially on unseen occlusions (29.67 vs 25.57 dB vs DeblurGAN-v2), depends on this hyperparameter, the reported margins are not a faithful estimate of generalization. Please pre-specify n, use a held-out validation split, or report results for a fixed n for all methods, together with variance.
[§2.3, Eq. (3)] The unified degradation model in Eq. (3) is presented as a derivation, but it is largely a restatement. Eq. (1) writes I = α·Is*h + (1−α)·Io*h, while Eq. (2) writes I ≈ Is*h + (α−1)·Ip*h + Ic*h; the relation between IN, IP, IC and the quantities in Eqs. (1)-(2) is not specified. The coefficient β is introduced without physical definition. The claim that partial occlusion removal is equivalent to defocus deblurring and complete occlusion removal to inpainting is asserted rather than proven. Please provide a rigorous derivation or reframe Eq. (3) as a notational convenience, and do not rely on it as a core theoretical contribution.
[Tables 1–3] All quantitative results are reported as single PSNR/SSIM values without error bars or statistical significance tests. With 333 test images, the margins over SOTA are sometimes small (e.g., Table 2, Dirt: PSNR 29.93 vs 28.52; Table 1, SSIM 0.879 vs 0.857) and may be within run-to-run variance. Please report means and standard deviations over multiple training runs or compute confidence intervals, and state the number of random seeds.
[§2.2] The dataset collection uses manual selection to ignore 'defective image pairs,' but the selection criteria and the number of discarded pairs are not reported. Residual misalignment between the contaminated and clean images, possibly due to refractive index mismatch or manual pairing, is not quantified. If the pairs are not pixel-aligned, the supervised loss trains on inconsistent targets and the reported PSNR gains could reflect alignment artifacts rather than true restoration. Please provide quantitative alignment statistics (e.g., residual displacement maps) or a sensitivity analysis on alignment.

minor comments (5)

[§4.2] The text says 'we train the model using Raindrops, Muddy Water, and Particles sub-datasets of the OSOR' — 'OSOR' should be 'OROS.'
[§4.3] The sentence 'The model is trained with various numbers of gradient updates' should read 'The model is adapted with various numbers of gradient updates,' since training is fixed and test-time adaptation varies.
[Eqs. (3), (7)] The symbol α is used both for the attenuation ratio in the degradation model and for the loss weight in the total loss. Please use different symbols to avoid confusion.
[Fig. 9] The caption says 'our method can utilize more frames in the testing time,' but there are no frames; the x-axis is the number of gradient updates. Please rephrase.
[§1, contributions] The phrase 'We will opensource our code and dataset as a promise' should be rephrased as a clear availability statement (e.g., 'We will release code and data upon publication').

Circularity Check

1 steps flagged · score 5.0 of 10

Test-time update count is tuned on the test set and the reported advantage is partly a selected maximum, not an independent prediction; otherwise the method's derivation is self-contained.

fitted input called prediction [Section 4.2 'Results on Unseen Ones' (Table 3), Section 4.3 'Number of Gradient Updates', Section 4.1 implementation details]
"We implement 20 updates and achieve the best quantitative performance in 3 updates. It validates the benefits of the test-time adaptation mechanism for better generalization. ... In our best case on OROS dataset, we set n as 4. ... We perform 6 gradient updates for each test image during the test-time adaptation."

The number of gradient updates n is selected on the test set by maximizing PSNR, and that selected value is then reported as the method's performance and used to claim generalization on unseen occlusions. For Table 3, 20 update counts are tried and the best (3 updates) is reported; for the main OROS comparison, n=4 is chosen because Fig. 9 shows it gives the best PSNR, while Section 4.1 states 6 updates. The reported PSNR is therefore the maximum over the tuned hyperparameter on the same evaluation set, so the 'prediction' is statistically forced by the selection rather than by a predetermined protocol. This inflates the apparent benefit of test-time adaptation and the margin over SOTA, though the base model without updates already beats the baselines.

full rationale

The core method—supervised training on real paired OROS data plus self-supervised test-time adaptation through an auxiliary reconstruction task—is not circular: the auxiliary loss uses only the degraded input as its target and the primary output is not defined in terms of the trainable parameters by construction. The unified degradation model (Eq. 3) is a loose re-labeling of unknown image components rather than a falsifiable derivation, but it is not used to generate numeric predictions, so it does not constitute circularity. The paper's self-citations (e.g., [18],[19]) are about autofocus/refractive-index issues and are not load-bearing for the method. The main circularity concern is evaluation: the test-time update count is tuned on the test set ('We implement 20 updates and achieve the best quantitative performance in 3 updates'; 'we get the best performance when gradient update times n = 4') and the reported PSNR is the selected maximum, while Section 4.1 says 6 updates. This makes the headline improvements partly a fitted-input-called-prediction artifact. Because even the no-update model beats the baselines, the central architectural claim retains independent content; hence score 5 rather than higher.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The paper does not introduce new physical entities. Its assumptions are standard domain models for image formation and self-supervised learning, plus an ad hoc decomposition in Eq. 3 that is not independently verified.

free parameters (3)

α0 = 0.9
Weight for the MS-SSIM term in the primary and auxiliary losses (Eq. 4-5); set empirically.
α = 0.8
Balance between primary and auxiliary loss in Eq. 7; set empirically.
n (test-time gradient steps) = 4 (selected by test-set performance)
Number of adaptation updates chosen from Fig. 9 to maximize test PSNR/SSIM, a form of tuning on the evaluation set.

assumptions (4)

domain assumption The degradation of an image through contaminated glass follows the image formation model I = α·Is*h + (1-α)·Io*h (Eq. 1)
Stated in Section 2.3 without validation; it motivates the network design.
ad hoc to paper The unified expression I = IN + (α-1)·IP + β·IC (Eq. 3) is a valid decomposition of the degraded image into non-occlusion, partial, and complete occlusion components.
The terms IN, IP, IC are not defined precisely; the equation is essentially a re-labeling that holds by construction, not a falsifiable physical law.
domain assumption The self-supervised auxiliary reconstruction task provides a useful learning signal for test-time adaptation.
The paper relies on this to justify updating the model on each test image; Section 3.1 assumes that reconstructing the degraded image teaches the model about the occlusion.
domain assumption The clean and degraded image pairs are pixel-aligned and the clean image is a reliable ground truth.
Manual selection and identical glass are used, but residual misalignment or environmental drift is not quantified (Section 2.2).

how reviews work

0 comments

Cite this review

Pith. "Pith review of Seeing through Unclear Glass: Occlusion Removal with One Shot." pith.science (2026). https://pith.science/paper/3224XVKD

@misc{pith2026250901033,
  author       = {Pith},
  title        = {Pith review of: Seeing through Unclear Glass: Occlusion Removal with One Shot},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3224XVKD}},
  note         = {Machine review of arXiv:2509.01033}
}

read the original abstract

Images taken through window glass are often degraded by contaminants adhered to the glass surfaces. Such contaminants cause occlusions that attenuate the incoming light and scatter stray light towards the camera. Most of existing deep learning methods for neutralizing the effects of contaminated glasses relied on synthetic training data. Few researchers used real degraded and clean image pairs, but they only considered removing or alleviating the effects of rain drops on glasses. This paper is concerned with the more challenging task of learning the restoration of images taken through glasses contaminated by a wide range of occluders, including muddy water, dirt and other small foreign particles found in reality. To facilitate the learning task we have gone to a great length to acquire real paired images with and without glass contaminants. More importantly, we propose an all-in-one model to neutralize contaminants of different types by utilizing the one-shot test-time adaptation mechanism. It involves a self-supervised auxiliary learning task to update the trained model for the unique occlusion type of each test image. Experimental results show that the proposed method outperforms the state-of-the-art methods quantitatively and qualitatively in cleaning realistic contaminated images, especially the unseen ones.

Figures

Figures reproduced from arXiv: 2509.01033 by the authors.

**Figure 1.** Occlusion removal of the unclear glass. We remove various occlusion types by the proposed all-in-one learning-based method with one shot. The occlusions contain (a) dirt, (b) raindrops, (c) muddy water, (d) particles. Our method achieves promising occlusion removal results. the problem by actively cleaning the glasses or retaking the images from different angles is impractical. With the emergence of deep learning i… view at source ↗

**Figure 2.** Illustration of three shooting scenarios.(a) Contaminants; (b) Obstructions; (c) Occlusions (Ours). 2.1. Differences of Shooting Scenarios In [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Models of occlusions.(a) Image formation model; (b) Defocus imaging model. where h(x, y, d) is the point spread function (PSF) at the defocusposition d for the occlusion layer, and ∗ denotes image convolution. The first term (attenuation) in Eq. 1 involves a convolution operator (pointwise multiplication), which allows the high-frequency components of Is to be partially preserved in the captured image I [7]. The s… view at source ↗

Figures from the paper (5 more)

**Figure 4.** Figure 4: Illustration of the proposed structure. The input of the network is an occlusion-degraded image I d ; the primary branch tries to recover a clean image I c ; the auxiliary branch tries to reconstruct the occlusion-degraded image ˆI d . We design an occlusion attenti…

**Figure 5.** Figure 5: Illustration of occlusion attention mask. all into the training data. Thus, we employ auxiliary learning, which effectively improves the performance of the primary task on unseen data [22]. In our scenarios, we choose self-supervised reconstruction to enforce the aux…

**Figure 6.** Figure 6: Qualitative comparison with state-of-the-art approaches on our OROS dataset. Our method yields sharper results than the state-of-the-art approaches. Methods Dataset OROS PSNR SSIM Degraded 23.31 0.846 PReNet [33] 22.80 0.835 U-net [41] 24.70 0.846 DeblurGAN-v2 [17] 28.…

**Figure 7.** Figure 7: Qualitative comparison with different network structures. (a) Primary; (b) Primary + Auxiliary; (c) Primary + Auxiliary + Update; (d) Primary + Auxiliary + Mask; (e) Primary + Auxiliary + Update + Mask. Network Structures PSNR SSIM Primary 29.16 0.869 Primary + Auxili…

**Figure 9.** Figure 9: Illustration of PSNR and SSIM after each gradient update. Our method can utilize more frames in the testing time for better performance until convergence. Note that we get the best performance when gradient update times n = 4. dates (n from 1 to 10). According to the …

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

50 extracted references · 43 canonical work pages

[1]

Muddy wa- ter animation with different details

Seungho Baek, Kiwon Um, and JungHyun Han. Muddy wa- ter animation with different details. Computer Animation and Virtual Worlds, 26(3-4):347–355, 2015. 3

work page 2015
[2]

Deep optics for monocu- lar depth estimation and 3d object detection

Julie Chang and Gordon Wetzstein. Deep optics for monocu- lar depth estimation and 3d object detection. In Proceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 10193–10202, 2019. 2

work page 2019
[3]

Test- time fast adaptation for dynamic scene deblurring via meta- auxiliary learning

Zhixiang Chi, Yang Wang, Yuanhao Yu, and Jin Tang. Test- time fast adaptation for dynamic scene deblurring via meta- auxiliary learning. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 9137–9146, 2021. 2, 3, 5

work page 2021
[4]

Cloud-RAIN: Point Cloud Analysis with Reflectional Invariance

Yiming Cui, Lecheng Ruan, Hang-Cheng Dong, Qiang Li, Zhongming Wu, Tieyong Zeng, and Feng-Lei Fan. Cloud- rain: point cloud analysis with reflectional invariance. arXiv preprint arXiv:2305.07814, 2023. 2

work page Pith review arXiv 2023
[5]

Lensfree auto-focusing imaging with coarse- to-fine tuning method

Zhihui Ding, Shenghao Zheng, Feilong Zhang, Qiang Li, and Cheng Guo. Lensfree auto-focusing imaging with coarse- to-fine tuning method. Optics and Lasers in Engineering , 181:108366, 2024. 2

work page 2024
[6]

Image super-resolution using deep convolutional net- works

Chao Dong, Chen Change Loy, Kaiming He, and Xiaoou Tang. Image super-resolution using deep convolutional net- works. IEEE transactions on pattern analysis and machine intelligence, 38(2):295–307, 2015. 1

work page 2015
[7]

Removing image artifacts due to dirty camera lenses and thin occluders

Jinwei Gu, Ravi Ramamoorthi, Peter Belhumeur, and Shree Nayar. Removing image artifacts due to dirty camera lenses and thin occluders. In ACM SIGGRAPH Asia 2009 papers , pages 1–10. 2009. 2, 4

work page 2009
[8]

A fast-converging iterative method based on weighted feedback for multi-distance phase retrieval

Cheng Guo, Cheng Shen, Qiang Li, Jiubin Tan, Shutian Liu, Xinchi Kan, and Zhengjun Liu. A fast-converging iterative method based on weighted feedback for multi-distance phase retrieval. Scientific reports, 8(1):6436, 2018. 2

work page 2018

Show all 50 references

[9]

Lensfree auto-focusing imaging using nuclear norm of gra- dient

Cheng Guo, Feilong Zhang, Xianming Liu, Qiang Li, Sheng- hao Zheng, Jiubin Tan, Zhengjun Liu, and Weibo Wang. Lensfree auto-focusing imaging using nuclear norm of gra- dient. Optics and lasers in engineering , 156:107076, 2022. 2

2022
[10]

Image inpaint- ing via conditional texture and structure dual generation

Xiefan Guo, Hongyu Yang, and Di Huang. Image inpaint- ing via conditional texture and structure dual generation. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 14134–14143, 2021. 6, 7

2021
[11]

Joint raindrop and haze removal from a single image

Yina Guo, Jianguo Chen, Xiaowen Ren, Anhong Wang, and Wenwu Wang. Joint raindrop and haze removal from a single image. IEEE Transactions on Image Processing , 29:9508– 9519, 2020. 2

2020
[12]

Deep residual learning for image recognition

Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceed- ings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016. 4

2016
[13]

Re- moval of image obstacles for vehicle-mounted surrounding monitoring cameras by real-time video inpainting

Yoshihiro Hirohashi, Kenichi Narioka, Masanori Suganuma, Xing Liu, Yukimasa Tamatsu, and Takayuki Okatani. Re- moval of image obstacles for vehicle-mounted surrounding monitoring cameras by real-time video inpainting. In Pro- ceedings of the IEEE/CVF Conference on Computer Vis...

2020
[14]

Amplitude-only, passive, broadband, optical spatial cloak- ing of very large objects

John C Howell, J Benjamin Howell, and Joseph S Choi. Amplitude-only, passive, broadband, optical spatial cloak- ing of very large objects. Applied Optics, 53(9):1958–1963,

1958
[15]

Reinforcement learning with unsupervised auxiliary tasks

Max Jaderberg, V olodymyr Mnih, Wojciech Marian Czar- necki, Tom Schaul, Joel Z Leibo, David Silver, and Koray Kavukcuoglu. Reinforcement learning with unsupervised auxiliary tasks. arXiv preprint arXiv:1611.05397, 2016. 2

2016 arXiv
[16]

Adam: A method for stochastic optimization

Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 ,

arXiv
[17]

Deblurgan-v2: Deblurring (orders-of-magnitude) faster and better

Orest Kupyn, Tetiana Martyniuk, Junru Wu, and Zhangyang Wang. Deblurgan-v2: Deblurring (orders-of-magnitude) faster and better. In Proceedings of the IEEE/CVF Inter- national Conference on Computer Vision, pages 8878–8887,
[18]

Learning to autofo- cus in whole slide imaging via physics-guided deep cascade networks

Qiang Li, Xianming Liu, Kaige Han, Cheng Guo, Junjun Jiang, Xiangyang Ji, and Xiaolin Wu. Learning to autofo- cus in whole slide imaging via physics-guided deep cascade networks. Optics Express, 30(9):14319–14340, 2022. 3

2022
[19]

Rapid whole slide imaging via dual-shot deep autofocusing

Qiang Li, Xianming Liu, Junjun Jiang, Cheng Guo, Xi- angyang Ji, and Xiaolin Wu. Rapid whole slide imaging via dual-shot deep autofocusing. IEEE Transactions on Compu- tational Imaging, 7:124–136, 2020. 2

2020
[20]

All in one bad weather removal using architectural search

Ruoteng Li, Robby T Tan, and Loong-Fah Cheong. All in one bad weather removal using architectural search. In Pro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 3175–3185, 2020. 2

2020
[21]

Let’s see clearly: Contaminant artifact removal for moving cam- eras

Xiaoyu Li, Bo Zhang, Jing Liao, and Pedro V Sander. Let’s see clearly: Contaminant artifact removal for moving cam- eras. In Proceedings of the IEEE/CVF International Con- ference on Computer Vision, pages 2011–2020, 2021. 1, 2, 3

2011
[22]

Self- supervised generalisation with meta auxiliary learning

Shikun Liu, Andrew Davison, and Edward Johns. Self- supervised generalisation with meta auxiliary learning. Ad- vances in Neural Information Processing Systems, 32, 2019. 5

2019
[23]

Learning deep models for face anti-spoofing: Binary or auxiliary su- pervision

Yaojie Liu, Amin Jourabloo, and Xiaoming Liu. Learning deep models for face anti-spoofing: Binary or auxiliary su- pervision. In Proceedings of the IEEE conference on com- puter vision and pattern recognition , pages 389–398, 2018. 2

2018
[24]

Learning to see through ob- structions

Yu-Lun Liu, Wei-Sheng Lai, Ming-Hsuan Yang, Yung-Yu Chuang, and Jia-Bin Huang. Learning to see through ob- structions. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 14215– 14224, 2020. 1, 2, 3, 5

2020
[25]

Removal of partial occlusion from single images

Scott McCloskey, Michael Langer, and Kaleem Siddiqi. Removal of partial occlusion from single images. IEEE transactions on pattern analysis and machine intelligence , 33(3):647–654, 2010. 2, 3

2010
[26]

Deep multi-scale convolutional neural network for dynamic scene deblurring

Seungjun Nah, Tae Hyun Kim, and Kyoung Mu Lee. Deep multi-scale convolutional neural network for dynamic scene deblurring. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 3883–3891,
[27]

Multi-temporal recurrent neural networks for progres- sive non-uniform single image deblurring with incremental temporal training

Dongwon Park, Dong Un Kang, Jisoo Kim, and Se Young Chun. Multi-temporal recurrent neural networks for progres- sive non-uniform single image deblurring with incremental temporal training. In European Conference on Computer Vi- sion, pages 327–343. Springer, 2020. 4

2020
[28]

Fast adaptation to super-resolution networks via meta-learning

Seobin Park, Jinsu Yoo, Donghyeon Cho, Jiwon Kim, and Tae Hyun Kim. Fast adaptation to super-resolution networks via meta-learning. In European Conference on Computer Vision, pages 754–769. Springer, 2020. 2

2020
[29]

Context encoders: Feature learning by inpainting

Deepak Pathak, Philipp Krahenbuhl, Jeff Donahue, Trevor Darrell, and Alexei A Efros. Context encoders: Feature learning by inpainting. In Proceedings of the IEEE con- ference on computer vision and pattern recognition , pages 2536–2544, 2016. 2, 7

2016
[30]

Attentive generative adversarial network for rain- drop removal from a single image

Rui Qian, Robby T Tan, Wenhan Yang, Jiajun Su, and Jiay- ing Liu. Attentive generative adversarial network for rain- drop removal from a single image. In Proceedings of the IEEE conference on computer vision and pattern recogni- tion, pages 2482–2491, 2018. 1, 2

2018
[31]

Remov- ing raindrops and rain streaks in one go

Ruijie Quan, Xin Yu, Yuanzhi Liang, and Yi Yang. Remov- ing raindrops and rain streaks in one go. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 9147–9156, 2021. 2

2021
[32]

Deep learning for seeing through window with raindrops

Yuhui Quan, Shijie Deng, Yixin Chen, and Hui Ji. Deep learning for seeing through window with raindrops. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 2463–2471, 2019. 2

2019
[33]

Progressive image deraining networks: A better and simpler baseline

Dongwei Ren, Wangmeng Zuo, Qinghua Hu, Pengfei Zhu, and Deyu Meng. Progressive image deraining networks: A better and simpler baseline. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 3937–3946, 2019. 6, 7

2019
[34]

Seeing through obstructions with diffractive cloaking

Zheng Shi, Yuval Bahat, Seung-Hwan Baek, Qiang Fu, Hadi Amata, Xiao Li, Praneeth Chakravarthula, Wolfgang Hei- drich, and Felix Heide. Seeing through obstructions with diffractive cloaking. ACM Transactions on Graphics (TOG), 41(4):1–15, 2022. 2, 5

2022
[35]

Test-time training with self- supervision for generalization under distribution shifts

Yu Sun, Xiaolong Wang, Zhuang Liu, John Miller, Alexei Efros, and Moritz Hardt. Test-time training with self- supervision for generalization under distribution shifts. In International conference on machine learning, pages 9229–
[36]

Reconstructing occluded sur- faces using synthetic apertures: Stereo, focus and robust measures

Vaibhav Vaish, Marc Levoy, Richard Szeliski, C Lawrence Zitnick, and Sing Bing Kang. Reconstructing occluded sur- faces using synthetic apertures: Stereo, focus and robust measures. In 2006 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR’06) ,...

2006
[37]

Deep auxiliary learning for visual localization and odometry

Abhinav Valada, Noha Radwan, and Wolfram Burgard. Deep auxiliary learning for visual localization and odometry. In 2018 IEEE international conference on robotics and au- tomation (ICRA), pages 6939–6946. IEEE, 2018. 2

2018
[38]

Image quality assessment: from error visibility to structural similarity

Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Si- moncelli. Image quality assessment: from error visibility to structural similarity. IEEE transactions on image processing, 13(4):600–612, 2004. 6

2004
[39]

High performance imaging us- ing large camera arrays

Bennett Wilburn, Neel Joshi, Vaibhav Vaish, Eino-Ville Tal- vala, Emilio Antunez, Adam Barth, Andrew Adams, Mark Horowitz, and Marc Levoy. High performance imaging us- ing large camera arrays. In ACM SIGGRAPH 2005 Papers, pages 765–776. 2005. 2

2005
[40]

Phase- cam3d—learning phase masks for passive single view depth estimation

Yicheng Wu, Vivek Boominathan, Huaijin Chen, Aswin Sankaranarayanan, and Ashok Veeraraghavan. Phase- cam3d—learning phase masks for passive single view depth estimation. In 2019 IEEE International Conference on Com- putational Photography (ICCP), pages 1–12. IEEE, 2019. 2

2019
[41]

Dense block u-net for dynamic scene deblur- ring

Yujie Wu, Hong Zhang, Yawei Li, Yinan Mao, Lei He, and Zhoufeng Liu. Dense block u-net for dynamic scene deblur- ring. In Proceedings of the 2020 The 4th International Con- ference on Graphics and Signal Processing , pages 46–50,

2020
[42]

Empirical evaluation of rectified activations in convolutional network

Bing Xu, Naiyan Wang, Tianqi Chen, and Mu Li. Empirical evaluation of rectified activations in convolutional network. arXiv preprint arXiv:1505.00853, 2015. 6

2015 arXiv
[43]

3dgs compression with sparsity-guided hierarchical transform coding

Hao Xu, Xiaolin Wu, and Xi Zhang. 3dgs compression with sparsity-guided hierarchical transform coding. arXiv preprint arXiv:2505.22908, 2025. 1

2025
[44]

Multirate neural im- age compression with adaptive lattice vector quantization

Hao Xu, Xiaolin Wu, and Xi Zhang. Multirate neural im- age compression with adaptive lattice vector quantization. In Proceedings of the Computer Vision and Pattern Recognition Conference (CVPR), pages 7633–7642, June 2025. 1

2025
[45]

Fast point cloud ge- ometry compression with context-based residual coding and inr-based refinement

Hao Xu, Xi Zhang, and Xiaolin Wu. Fast point cloud ge- ometry compression with context-based residual coding and inr-based refinement. In European Conference on Computer Vision, pages 270–288. Springer, 2024. 1

2024
[46]

Self-supervised scene de- occlusion

Xiaohang Zhan, Xingang Pan, Bo Dai, Ziwei Liu, Dahua Lin, and Chen Change Loy. Self-supervised scene de- occlusion. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 3784– 3792, 2020. 3

2020
[47]

Dual attention-in-attention model for joint rain streak and raindrop removal

Kaihao Zhang, Dongxu Li, Wenhan Luo, and Wenqi Ren. Dual attention-in-attention model for joint rain streak and raindrop removal. IEEE Transactions on Image Processing, 30:7608–7619, 2021. 1

2021
[48]

Beyond a gaussian denoiser: Residual learning of deep cnn for image denoising

Kai Zhang, Wangmeng Zuo, Yunjin Chen, Deyu Meng, and Lei Zhang. Beyond a gaussian denoiser: Residual learning of deep cnn for image denoising. IEEE transactions on image processing, 26(7):3142–3155, 2017. 1

2017
[49]

Loss functions for image restoration with neural networks

Hang Zhao, Orazio Gallo, Iuri Frosio, and Jan Kautz. Loss functions for image restoration with neural networks. IEEE Transactions on computational imaging , 3(1):47–57, 2016. 5

2016
[50]

Unsupervised learning of depth and ego-motion from video

Tinghui Zhou, Matthew Brown, Noah Snavely, and David G Lowe. Unsupervised learning of depth and ego-motion from video. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 1851–1858, 2017. 2

2017

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.

[1] [1]

Muddy wa- ter animation with different details

Seungho Baek, Kiwon Um, and JungHyun Han. Muddy wa- ter animation with different details. Computer Animation and Virtual Worlds, 26(3-4):347–355, 2015. 3

work page 2015

[2] [2]

Deep optics for monocu- lar depth estimation and 3d object detection

Julie Chang and Gordon Wetzstein. Deep optics for monocu- lar depth estimation and 3d object detection. In Proceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 10193–10202, 2019. 2

work page 2019

[3] [3]

Test- time fast adaptation for dynamic scene deblurring via meta- auxiliary learning

Zhixiang Chi, Yang Wang, Yuanhao Yu, and Jin Tang. Test- time fast adaptation for dynamic scene deblurring via meta- auxiliary learning. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 9137–9146, 2021. 2, 3, 5

work page 2021

[4] [4]

Cloud-RAIN: Point Cloud Analysis with Reflectional Invariance

Yiming Cui, Lecheng Ruan, Hang-Cheng Dong, Qiang Li, Zhongming Wu, Tieyong Zeng, and Feng-Lei Fan. Cloud- rain: point cloud analysis with reflectional invariance. arXiv preprint arXiv:2305.07814, 2023. 2

work page Pith review arXiv 2023

[5] [5]

Lensfree auto-focusing imaging with coarse- to-fine tuning method

Zhihui Ding, Shenghao Zheng, Feilong Zhang, Qiang Li, and Cheng Guo. Lensfree auto-focusing imaging with coarse- to-fine tuning method. Optics and Lasers in Engineering , 181:108366, 2024. 2

work page 2024

[6] [6]

Image super-resolution using deep convolutional net- works

Chao Dong, Chen Change Loy, Kaiming He, and Xiaoou Tang. Image super-resolution using deep convolutional net- works. IEEE transactions on pattern analysis and machine intelligence, 38(2):295–307, 2015. 1

work page 2015

[7] [7]

Removing image artifacts due to dirty camera lenses and thin occluders

Jinwei Gu, Ravi Ramamoorthi, Peter Belhumeur, and Shree Nayar. Removing image artifacts due to dirty camera lenses and thin occluders. In ACM SIGGRAPH Asia 2009 papers , pages 1–10. 2009. 2, 4

work page 2009

[8] [8]

A fast-converging iterative method based on weighted feedback for multi-distance phase retrieval

Cheng Guo, Cheng Shen, Qiang Li, Jiubin Tan, Shutian Liu, Xinchi Kan, and Zhengjun Liu. A fast-converging iterative method based on weighted feedback for multi-distance phase retrieval. Scientific reports, 8(1):6436, 2018. 2

work page 2018